Development of a controller to switch between relative and absolute path for target vehicles in simulation scenarios

Detta är en Uppsats för yrkesexamina på avancerad nivå från Lunds universitet/Institutionen för reglerteknik

Författare: Olof Karlsson; Erik Fredin; [2021]

Nyckelord: Technology and Engineering;

Sammanfattning: Simulating vehicle scenarios for autonomous cars saves both time and money compared to physical testing. For this to work efficiently a sophisticated simulation software is often needed. Esmini is an open-source, lightweight vehicle simulator that can handle advanced traffic scenarios. When designing scenarios to be used for testing autonomous vehicles, it is often beneficial to set up relative vehicle behaviour to ensure that a hazardous situation arises. This could be that the car in front of the autonomous car drives a couple of kilometers per hour slower compared to the autonomous car. At some point the autonomous car must react to avoid a collision, e.g., by braking. To avoid having the car in front slowing down when the autonomous car brakes, a switch must be done such that the car in front has an absolute speed. Hence, the purpose of the controller developed in this thesis is to decide when to switch from the relative setup and what the new control objective should be. This is implemented by simulating a short period into the future to observe how the scenario develops. The controller generates predictions using simulation data of the events that should transpire for every vehicle except the autonomous one. The expected vehicle states are then continuously compared to the actual vehicle states generated during the simulation. If the actual vehicle states deviate too far from the expected values, a switch occurs. A new objective is then given to the vehicle, depending on its behaviour before the switch and what type of relative setup previously existed. The proposed solution is available in the official Esmini repository and has been tested in traffic scenarios with CSPAS, a platform where Volvo’s active safety functions are simulated. The performance of the controller depends on how it is tuned. For all tested scenarios a satisfying result has been achieved.

  HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)